Learning plateaus

Why does progress in a new skill stall for weeks and then jump without extra effort?

The shape is familiar to anyone who has taken up an instrument, a language or a sport. Rapid gains for a while, then a flat stretch that outlasts your patience, and then — often after a week when you did less, not more — a jump. Nothing in the practice changed to explain it.

The everyday account is diminishing returns: you are approaching your limit, so progress slows. But look at the shape that theory predicts. Diminishing returns gives a curve that bends gently and keeps rising; a hard limit gives a curve that flattens and stays flat. Neither of them gives flat-then-jump. So whatever produces a plateau, it is not simply the top of the curve arriving.

Ask a narrower question: what could hold a measured score still while learning continued underneath it?

The first possibility is the dullest and the most common. The number you are watching is one dimension of a performance that has several. A typist stuck at the same words per minute may be making fewer errors, recovering from them faster, and finding the work less exhausting. A climber stuck on the same grade may be doing it with far less panic. All of that is learning, and none of it moves the metric you happen to be scoring. Some plateaus are entirely artefacts of what is being counted.

The second is more interesting. Suppose the skill is being performed by a particular method, and that method has a ceiling. Practice can polish the method towards its ceiling and no further, however diligently applied. Progress then resumes only when the unit changes — when what used to be several actions becomes one. Bryan and Harter observed exactly this in 1897 in telegraph operators, who seemed to move from handling letters to handling whole words, and their plateau has been the textbook case ever since.

It is worth being careful here, because the textbook case is contested. In 1958 Keller published "The phantom plateau", arguing that the plateau was not an inevitable stage of skill but a product of how the operators had been taught, and pointing to learners trained differently who showed none. The safest reading is that plateaus are a real and common experience, but not a law — some are structural, and some are training defects wearing the costume of a law.

Third, why would the change of unit arrive suddenly? Because it depends on a condition being met rather than on effort accumulating. A larger unit can only form once the smaller components run without attention, and attention is the scarce resource: while you are still spending it on individual letters there is none left to notice the shape of a word. Automaticity develops gradually, but its consequence — the freeing of enough capacity for a reorganisation — crosses a threshold. Gradual cause, abrupt effect.

And one more thing, about the smooth learning curve you have seen in books. It is nearly always an average over many learners. If individuals step upward at different moments, averaging their steps produces a graceful ramp that resembles none of them. Whether individual learning is fundamentally smooth or fundamentally stepwise is still argued, but the aggregation point stands on its own: the tidy curve is a property of the group.

As for the jump arriving "without extra effort" — the effort went in during the flat stretch, where it was invisible because it was going into the reorganisation rather than into the score. Rest between sessions genuinely contributes too, but the plateau is not idle time; it is the time the machinery was being rebuilt.

A plateau is not one thing, and the useful move is to ask which kind you are on. If the gains have moved to a dimension you are not scoring, nothing is wrong and the fix is to measure something else. If the current method has hit its ceiling, more repetitions of it are pure cost — what breaks the stall is changing what you practise: isolating a sub-skill, imposing a constraint that makes the old method impossible, or getting feedback on a dimension you had not been attending to. If the reorganisation simply has not finished yet, the practice is working and the score is lagging.

Two honest qualifications. It is often impossible to tell these apart from inside, which is a large part of why plateaus are demoralising. And some plateaus really are ceilings — of the equipment, the time available, or the person — and telling a genuine ceiling from a gear that has not been changed is a judgement, not a calculation.

• How to detect from inside which kind of plateau you are on, before spending months on the wrong response.
• Why deliberate constraints — slower tempo, weaker hand, restricted vocabulary — so often break a stall.
• Whether individual learning curves are genuinely stepwise or the steps are an artefact of coarse measurement.

Every term the collection defines is gathered in the glossary.